Interactive Project
Ballmatics Academy wanted its basketball and math program to work at home as well as it does at its downtown Toronto facility, keeping kids engaged without a coach in the room. Ollon built an interactive layer into the mobile app where basketball activity, detected through the camera, feeds directly into math practice and testing, so a made basket becomes part of the lesson itself. The scoring system reads different states from the detection model, tracking whether a shot connects, and turns that into points and progress inside the app. Kids work through self directed lessons paired with study buddies, keeping the experience social even when they are not in the same room.
Machine Learning
Detecting a basketball in real time on a phone camera, accurately enough that the results could power the app's games and lessons directly, was the core technical problem Ollon solved for Ballmatics Academy. Ollon trained custom YOLO based detection models on a purpose built dataset of basketball footage covering shot detection and sub region analysis. To run on mobile hardware, models were converted from PyTorch to TorchScript and then to the NCNN format, with OpenCV handling image processing in the native Android layer. An automated pipeline tracked training sets, classifiers, and test runs in a database, comparing accuracy across model versions before choosing which one shipped.
Mobile Application Development
Ballmatics Academy needed a mobile app that could run a real time detection model on a phone while still feeling like a simple app for kids to open and play. Ollon built the application in React Native with a NodeJS backend, and added a native Android layer in C++ to run the trained detection model directly on device through JNI. One packaging bug in the debug build left screens blank until the team switched build types to isolate it. That native detection layer is what let the app run ball tracking in real time without noticeable lag, giving Ballmatics Academy a functional mobile prototype ready for further testing.
MVP development and product-market fit
Ballmatics Academy set out to bring its in person basketball and math program into a testable mobile prototype before considering a wider release. Ollon built that prototype to a functional state, combining a working detection model, interactive games, and lesson content on a real device, and kept it moving through an active cycle of staging and release builds. Testing covered build stability, detection accuracy across model versions, and smoke testing new releases before they went out for further evaluation. That functional prototype gave Ballmatics Academy something concrete to test with users and content developers as the program moved toward a wider release.
